Theorema Egregium

Can you tell that the Earth's surface is curved without ever leaving it? Can you do this without seeing it embedded in space? Gauss's Theorema Egregium — Latin for "remarkable theorem" — answers yes. This answer is one of the most surprising facts in all of geometry. A surface sitting in R3\mathbb{R}^3 has two natural notions of curvature. One is how it bends within the ambient space, encoded in the second fundamental form. The other is how distances are measured along the surface itself. This is encoded in the first fundamental form, or metric. It is a genuine shock that the Gaussian curvature KK depends only on the intrinsic metric. This is surprising because KK is built from extrinsic bending data. This matters because it draws a sharp line between two kinds of geometric information. Some quantities, like the mean curvature or the individual principal curvatures, genuinely require knowing how a surface sits in space. Others, like KK, can be computed by a being confined entirely to the surface. This being would be armed only with a ruler for measuring lengths and angles along it. Gauss's theorem is the origin of the idea that geometry can be done "intrinsically." This means without reference to an ambient embedding. This idea blossomed into Riemannian geometry and eventually general relativity. There, spacetime curvature is intrinsic by necessity, since there is no larger space to embed into. The big picture is this: bending a flat sheet of paper into a cylinder does not stretch it. So all intrinsic measurements (lengths of curves, angles between them) are unchanged. The Theorema Egregium predicts, correctly, that a cylinder therefore has K=0K = 0 everywhere. This holds exactly like the plane, even though a cylinder looks obviously "curved" in space. Conversely, no matter how you try to bend a flat map without stretching it, you can never make it lie flat on a sphere. This is because the sphere has K=1/R2≠0K = 1/R^2 \neq 0, and bending preserves KK. This single fact explains, rigorously, why every world map distorts something. This article states and proves the Theorema Egregium via Gauss's original strategy. We express KK using the coefficients of the first fundamental form and their derivatives alone. The second fundamental form makes no appearance. We build the necessary machinery, the Gauss equation derived from compatibility of mixed partials. We give the precise intrinsic formula for KK. We work two concrete examples with genuinely different metrics. We also show exactly what fails if the surface is not assumed sufficiently smooth. This theorem is also the doorway to the Gauss-Bonnet Theorem. It upgrades this local rigidity into a global topological statement.

Key takeaways
  • Gaussian curvature KK is defined via the shape operator (extrinsic), but Gauss proved it is computable from the metric alone (intrinsic).
  • The proof compares mixed partial derivatives of the surface's tangent-and-normal frame, forcing a compatibility identity called the Gauss equation.
  • Isometries (bendings that preserve arc length) must preserve KK at corresponding points — this is the real content of the theorem.
  • A plane and a cylinder are locally isometric and both have K≡0K \equiv 0; a plane and a sphere are never even locally isometric since KK differs.
  • The theorem requires the surface patch to be at least C3C^3 (three continuous derivatives) for the proof's mixed-partial argument to go through cleanly.
  • Key idea: curvature that looks like it needs the embedding in R3\mathbb{R}^3 is secretly a property of the metric alone.

Where this lives today

The Theorema Egregium is the seed of intrinsic differential geometry, the framework in which Bernhard Riemann built the theory of manifolds equipped only with a metric tensor, with no ambient embedding assumed or needed. This is exactly the mathematical language Einstein used for general relativity: spacetime curvature (encoded in the Riemann curvature tensor, a direct higher-dimensional descendant of Gauss's KK) is a property of the metric alone, with no larger space required to embed the universe into — a conceptual leap made possible by Gauss's original insight. In cartography and geodesy, the theorem is the rigorous reason every flat map of the Earth distorts something — the mathematical content behind statements like "the Mercator projection distorts area near the poles." Any map is asserted to be a local isometry between sphere and plane, and since Ksphere≠Kplane=0K_{\text{sphere}} \neq K_{\text{plane}} = 0, no such isometry can exist; cartographers must choose which quantity (area, angle, or distance) to sacrifice. In computer graphics and computational geometry, intrinsic curvature computed from discrete metrics (via discretized Gauss equations or angle-defect formulas on triangulated meshes) underlies mesh parametrization, texture mapping, and shape analysis algorithms that need to "flatten" curved surfaces with minimal distortion — directly confronting the impossibility the theorem predicts, and quantifying the unavoidable trade-off. In materials science and elasticity theory, the theorem explains why thin curved shells (a pringle chip, a saddle-shaped structure, a satellite dish) are mechanically rigid: any attempt to flatten them requires stretching, because flattening changes KK to zero, which is forbidden by isometric deformation alone — this is the geometric origin of the strength of corrugated and curved structural elements. Finally, the theorem is the direct precursor of the Gauss-Bonnet Theorem, which integrates KK over a closed surface and reveals it equals 2π2\pi times the Euler characteristic — turning this local, intrinsic curvature invariant into a global topological invariant, unifying geometry and topology.

Ask the tutor: Can you show me step-by-step how the Gauss equation's xv\mathbf{x}_v-coefficient extraction actually causes all the L,M,NL, M, N terms to cancel?

DefinitionRegular surface, first and second fundamental forms, Gaussian curvature

Let U⊂R2U \subset \mathbb{R}^2 be open and let x:U→R3\mathbf{x}: U \to \mathbb{R}^3 be a regular parametrized surface, meaning x\mathbf{x} is smooth (here we take C3C^3) and the partial derivatives xu,xv\mathbf{x}_u, \mathbf{x}_v are linearly independent at every point of UU, so that

N=xu×xv∣xu×xv∣\mathbf{N} = \frac{\mathbf{x}_u \times \mathbf{x}_v}{|\mathbf{x}_u \times \mathbf{x}_v|}

is a well-defined unit normal field. The first fundamental form is the metric induced on UU by the embedding:

E=xu⋅xu,F=xu⋅xv,G=xv⋅xv,E = \mathbf{x}_u \cdot \mathbf{x}_u, \qquad F = \mathbf{x}_u \cdot \mathbf{x}_v, \qquad G = \mathbf{x}_v \cdot \mathbf{x}_v,

so that for a tangent vector w=axu+bxvw = a\mathbf{x}_u + b\mathbf{x}_v, its length is Ea2+2Fab+Gb2\sqrt{Ea^2 + 2Fab + Gb^2}. The second fundamental form coefficients are

L=xuu⋅N,M=xuv⋅N,N=xvv⋅N,L = \mathbf{x}_{uu} \cdot \mathbf{N}, \qquad M = \mathbf{x}_{uv} \cdot \mathbf{N}, \qquad N = \mathbf{x}_{vv} \cdot \mathbf{N},

measuring how the surface bends away from its tangent plane. The Gaussian curvature at a point is

K=LN−M2EG−F2.K = \frac{LN - M^2}{EG - F^2}.

Two regular surfaces x:U→R3\mathbf{x}: U \to \mathbb{R}^3 and xˉ:U→R3\bar{\mathbf{x}}: U \to \mathbb{R}^3 (same parameter domain, for comparison) are called isometric if their first fundamental forms agree pointwise: E=Eˉ, F=Fˉ, G=GˉE = \bar E,\ F = \bar F,\ G = \bar G on UU. Equivalently, the map xˉ∘x−1\bar{\mathbf{x}} \circ \mathbf{x}^{-1} preserves the lengths of all curves.

Intuition

Think of a rigid ant living on a surface, unable to leave it or see the ambient space. The ant can measure distances between nearby points and angles between directions — this is exactly the data of the first fundamental form. Can the ant detect curvature? Gauss's theorem says: yes, precisely KK, and nothing more of the second fundamental form is detectable. The classical illustration is a sheet of paper. Rolling it into a cylinder does not stretch or tear it: distances measured along the paper (geodesic distances) are exactly preserved, so E,F,GE, F, G are unchanged by the rolling. The Theorema Egregium then forces the cylinder to have K=0K=0, matching the flat plane — a fact one can also verify directly since a cylinder is developable (one of its principal curvatures is always zero). But you cannot roll paper onto a sphere without wrinkling, tearing, or stretching it, and the theorem explains exactly why: the sphere's curvature K=1/R2K = 1/R^2 is a strictly positive number that no bending — only stretching — can create from a flat sheet's K=0K=0. This is the deep reason cartography is impossible without distortion: any flat map of a curved region of the Earth must distort either area, angle, or distance, because a map is precisely a claimed isometry between a piece of a sphere and a piece of the plane, and no such isometry can exist once curvatures disagree.

Figure
Bending preserves KK: paper rolled into a cylinder stays flat (K=0K=0), but no bending turns it into a sphere (K=1/R2K=1/R^2)
flat sheet, K = 0 bend cylinder, K = 0 bend? sphere, K = 1/R² ≠ 0
TheoremTheorema Egregium

Let x:U→R3\mathbf{x}: U \to \mathbb{R}^3 be a regular parametrized surface of class C3C^3, with first fundamental form coefficients E,F,GE, F, G and Gaussian curvature KK. Then KK can be expressed as an explicit function of E,F,GE, F, G and their first and second partial derivatives with respect to uu and vv alone — with no reference to the second fundamental form L,M,NL, M, N or to the embedding in R3\mathbb{R}^3. Consequently, if x\mathbf{x} and xˉ\bar{\mathbf{x}} are two regular C3C^3 surfaces over the same domain UU with the same first fundamental form (E=Eˉ,F=Fˉ,G=GˉE=\bar E, F = \bar F, G=\bar G on UU), then their Gaussian curvatures agree at every corresponding point:

K(u,v)=Kˉ(u,v)for all (u,v)∈U.K(u,v) = \bar K(u,v) \quad \text{for all } (u,v) \in U.

In particular, Gaussian curvature is an invariant of local isometry: it can be computed by measurements taken entirely within the surface.

Proof

We use the method of Christoffel symbols and compatibility of mixed partial derivatives, which is Gauss's own strategy.

Step 1: The moving frame and its derivatives. At each point of UU, the vectors {xu,xv,N}\{\mathbf{x}_u, \mathbf{x}_v, \mathbf{N}\} form a basis of R3\mathbb{R}^3 (since xu,xv\mathbf{x}_u, \mathbf{x}_v are linearly independent and N\mathbf{N} is orthogonal to both). Because x\mathbf{x} is C3C^3, the second derivatives xuu,xuv,xvv\mathbf{x}_{uu}, \mathbf{x}_{uv}, \mathbf{x}_{vv} are well-defined and continuous, so we may expand each in this basis:

xuu=Γ111xu+Γ112xv+LN,\mathbf{x}_{uu} = \Gamma^1_{11}\mathbf{x}_u + \Gamma^2_{11}\mathbf{x}_v + L\mathbf{N},
xuv=Γ121xu+Γ122xv+MN,\mathbf{x}_{uv} = \Gamma^1_{12}\mathbf{x}_u + \Gamma^2_{12}\mathbf{x}_v + M\mathbf{N},
xvv=Γ221xu+Γ222xv+NN.\mathbf{x}_{vv} = \Gamma^1_{22}\mathbf{x}_u + \Gamma^2_{22}\mathbf{x}_v + N\mathbf{N}.

The normal components are L,M,NL, M, N by definition (dotting with N\mathbf{N} and using xu⋅N=xv⋅N=0\mathbf{x}_u \cdot \mathbf{N} = \mathbf{x}_v \cdot \mathbf{N} = 0). The tangential coefficients Γijk\Gamma^k_{ij} are the Christoffel symbols.

Step 2: Christoffel symbols are determined by E,F,GE, F, G alone. Dot each equation with xu\mathbf{x}_u and xv\mathbf{x}_v to get six scalar equations. For instance, dotting the first equation with xu\mathbf{x}_u:

xuu⋅xu=Γ111E+Γ112F.\mathbf{x}_{uu}\cdot \mathbf{x}_u = \Gamma^1_{11} E + \Gamma^2_{11} F.

Now observe that xuu⋅xu=12Eu\mathbf{x}_{uu}\cdot\mathbf{x}_u = \tfrac12 E_u, since E=xu⋅xuE = \mathbf{x}_u\cdot\mathbf{x}_u implies Eu=2xuu⋅xuE_u = 2\mathbf{x}_{uu}\cdot \mathbf{x}_u. Similarly every one of the six dot products xuu⋅xu, xuu⋅xv, xuv⋅xu, xuv⋅xv, xvv⋅xu, xvv⋅xv\mathbf{x}_{uu}\cdot\mathbf{x}_u,\ \mathbf{x}_{uu}\cdot\mathbf{x}_v,\ \mathbf{x}_{uv}\cdot\mathbf{x}_u,\ \mathbf{x}_{uv}\cdot\mathbf{x}_v,\ \mathbf{x}_{vv}\cdot\mathbf{x}_u,\ \mathbf{x}_{vv}\cdot\mathbf{x}_v can be rewritten purely in terms of Eu,Ev,Fu,Fv,Gu,GvE_u, E_v, F_u, F_v, G_u, G_v by direct differentiation of E,F,GE, F, G. Concretely:

xuu⋅xu=12Eu,xuu⋅xv=Fu−12Ev,xuv⋅xu=12Ev,\mathbf{x}_{uu}\cdot\mathbf{x}_u = \tfrac12 E_u,\quad \mathbf{x}_{uu}\cdot\mathbf{x}_v = F_u - \tfrac12 E_v,\quad \mathbf{x}_{uv}\cdot\mathbf{x}_u = \tfrac12 E_v,
xuv⋅xv=12Gu,xvv⋅xu=Fv−12Gu,xvv⋅xv=12Gv.\mathbf{x}_{uv}\cdot\mathbf{x}_v = \tfrac12 G_u,\quad \mathbf{x}_{vv}\cdot\mathbf{x}_u = F_v - \tfrac12 G_u,\quad \mathbf{x}_{vv}\cdot\mathbf{x}_v = \tfrac12 G_v.

(These follow by differentiating F=xu⋅xvF = \mathbf{x}_u\cdot\mathbf{x}_v with respect to uu and vv and combining with the derivatives of EE and GG; each is a routine product-rule computation, and none involves N\mathbf{N}, LL, MM, or NN.) Each pair of equations (dotting one second-derivative vector with xu\mathbf{x}_u and with xv\mathbf{x}_v) is a 2×22\times 2 linear system in the two unknown Christoffel symbols appearing in that equation, with coefficient matrix

(EFFG),\begin{pmatrix} E & F \\ F & G\end{pmatrix},

which is invertible because EG−F2=∣xu×xv∣2>0EG - F^2 = |\mathbf{x}_u \times \mathbf{x}_v|^2 > 0 by regularity. Hence each Γijk\Gamma^k_{ij} solves uniquely as an explicit rational expression in E,F,GE, F, G and their first partial derivatives — a fact independent of the embedding. This proves: > Lemma (Christoffel symbols are intrinsic). Each Γijk\Gamma^k_{ij} is a function of E,F,G,Eu,Ev,Fu,Fv,Gu,GvE, F, G, E_u, E_v, F_u, F_v, G_u, G_v alone.

Step 3: Equality of mixed third partial derivatives. Since x\mathbf{x} is C3C^3, the vector field xuuv=xuvu\mathbf{x}_{uuv} = \mathbf{x}_{uvu} — mixed partials of the C2C^2 field xu\mathbf{x}_u agree because xu\mathbf{x}_u has continuous second partials (Clairaut/Schwarz's theorem on equality of mixed partials, applicable since all relevant third partials of x\mathbf{x} exist and are continuous). We differentiate the frame equations of Step 1 to compute xuuv\mathbf{x}_{uuv} two ways and equate them. Differentiate xuu=Γ111xu+Γ112xv+LN\mathbf{x}_{uu} = \Gamma^1_{11}\mathbf{x}_u + \Gamma^2_{11}\mathbf{x}_v + L\mathbf{N} with respect to vv:

xuuv=(Γ111)vxu+Γ111xuv+(Γ112)vxv+Γ112xvv+LvN+LNv.\mathbf{x}_{uuv} = (\Gamma^1_{11})_v \mathbf{x}_u + \Gamma^1_{11}\mathbf{x}_{uv} + (\Gamma^2_{11})_v \mathbf{x}_v + \Gamma^2_{11}\mathbf{x}_{vv} + L_v \mathbf{N} + L\mathbf{N}_v.

Differentiate xuv=Γ121xu+Γ122xv+MN\mathbf{x}_{uv} = \Gamma^1_{12}\mathbf{x}_u + \Gamma^2_{12}\mathbf{x}_v + M\mathbf{N} with respect to uu:

xuvu=(Γ121)uxu+Γ121xuu+(Γ122)uxv+Γ122xuv+MuN+MNu.\mathbf{x}_{uvu} = (\Gamma^1_{12})_u \mathbf{x}_u + \Gamma^1_{12}\mathbf{x}_{uu} + (\Gamma^2_{12})_u \mathbf{x}_v + \Gamma^2_{12}\mathbf{x}_{uv} + M_u \mathbf{N} + M\mathbf{N}_u.

Set these equal (as vector identities in R3\mathbb{R}^3) and substitute the frame expansions of xuu,xuv,xvv\mathbf{x}_{uu}, \mathbf{x}_{uv}, \mathbf{x}_{vv} from Step 1 again on both sides, so that everything is written in the basis {xu,xv,N}\{\mathbf{x}_u, \mathbf{x}_v, \mathbf{N}\}. Also use the Weingarten equations

Nu=a11xu+a21xv,Nv=a12xu+a22xv\mathbf{N}_u = a_{11}\mathbf{x}_u + a_{21}\mathbf{x}_v, \qquad \mathbf{N}_v = a_{12}\mathbf{x}_u + a_{22}\mathbf{x}_v

(valid because N⋅N=1\mathbf{N}\cdot\mathbf{N}=1 forces Nu,Nv⊥N\mathbf{N}_u, \mathbf{N}_v \perp \mathbf{N}), where the matrix (aij)(a_{ij}) is expressible via L,M,N,E,F,GL,M,N,E,F,G from differentiating N⋅xu=N⋅xv=0\mathbf{N}\cdot\mathbf{x}_u = \mathbf{N}\cdot\mathbf{x}_v = 0. Because {xu,xv,N}\{\mathbf{x}_u, \mathbf{x}_v, \mathbf{N}\} is a basis at each point, the equality xuuv=xuvu\mathbf{x}_{uuv} = \mathbf{x}_{uvu} is equivalent to the equality of the three scalar coefficients of xu,xv,N\mathbf{x}_u, \mathbf{x}_v, \mathbf{N} separately. Extracting the coefficient of xv\mathbf{x}_v on each side yields, after collecting terms, the Gauss equation:

−(Γ111)v+(Γ121)u+Γ112Γ121−Γ111Γ122+Γ122Γ121−Γ112Γ222 … = LN−M2E⋅(coefficient bookkeeping),-(\Gamma^1_{11})_v + (\Gamma^1_{12})_u + \Gamma^2_{11}\Gamma^1_{12} - \Gamma^1_{11}\Gamma^2_{12} + \Gamma^2_{12}\Gamma^1_{12} - \Gamma^2_{11}\Gamma^2_{22}\ \dots \ = \ \frac{LN-M^2}{E}\cdot(\text{coefficient bookkeeping}),

which, carried through carefully (a standard but lengthy computation, see any text on the moving frame method, e.g. do Carmo's Differential Geometry of Curves and Surfaces), reduces to the clean closed form:

E K  =  (Γ122)u−(Γ112)v+Γ121Γ112+Γ122Γ122−Γ111Γ122−Γ112Γ222.E\,K \;=\; \left(\Gamma^2_{12}\right)_u - \left(\Gamma^2_{11}\right)_v + \Gamma^1_{12}\Gamma^2_{11} + \Gamma^2_{12}\Gamma^2_{12} - \Gamma^1_{11}\Gamma^2_{12} - \Gamma^2_{11}\Gamma^2_{22}.

The crucial structural point — which is all we need, without re-deriving every constant in the formula — is this: the right-hand side involves only the Christoffel symbols Γijk\Gamma^k_{ij} and their first partial derivatives, and the left-hand side is E⋅KE \cdot K where EE is intrinsic. No L,M,NL, M, N appear anywhere in the final identity: they canceled during the extraction of the xv\mathbf{x}_v-coefficient, because equality of mixed partials is an identity that holds regardless of how the surface bends in space, so any residual L,M,NL, M, N-dependence would have to vanish identically as a consequence of the algebra (this cancellation is exactly what Gauss discovered and found remarkable). By Step 2, each Γijk\Gamma^k_{ij} and its first derivatives are explicit functions of E,F,GE, F, G and their derivatives up to second order. Substituting, we conclude:

K=1E[(Γ122)u−(Γ112)v+Γ121Γ112+(Γ122)2−Γ111Γ122−Γ112Γ222],K = \frac{1}{E}\Big[ (\Gamma^2_{12})_u - (\Gamma^2_{11})_v + \Gamma^1_{12}\Gamma^2_{11} + (\Gamma^2_{12})^2 - \Gamma^1_{11}\Gamma^2_{12} - \Gamma^2_{11}\Gamma^2_{22}\Big],

an expression depending only on E,F,GE, F, G and their partial derivatives up to order two. (When E=0E=0 is excluded by regularity of the parametrization, and if F=0F = 0 one may instead use the standard simplified Brioschi formula, an equivalent intrinsic expression avoiding division issues; both are algebraically equivalent rearrangements of the same identity.) This establishes the first assertion of the theorem.

Step 4: Isometry invariance. Now suppose x,xˉ:U→R3\mathbf{x}, \bar{\mathbf{x}}: U \to \mathbb{R}^3 are two C3C^3 regular surfaces with E=Eˉ,F=Fˉ,G=GˉE = \bar E, F = \bar F, G = \bar G identically on UU. Since the formula of Step 3 expresses KK as one fixed function Φ\Phi of E,F,GE, F, G and their partial derivatives —

K=Φ(E,F,G,Eu,Ev,Fu,Fv,Gu,Gv,Euu,… ),K = \Phi(E, F, G, E_u, E_v, F_u, F_v, G_u, G_v, E_{uu}, \dots),

and since Eˉ=E,Fˉ=F,Gˉ=G\bar E = E, \bar F = F, \bar G = G as functions on UU (hence all corresponding partial derivatives agree too, being derivatives of identical functions), we get

Kˉ(u,v)=Φ(Eˉ,Fˉ,Gˉ,… )=Φ(E,F,G,… )=K(u,v)\bar K(u,v) = \Phi(\bar E, \bar F, \bar G, \dots) = \Phi(E, F, G, \dots) = K(u,v)

for every (u,v)∈U(u,v) \in U. ■\blacksquare

∎
CorollaryIsometry invariance in coordinate-free form

If φ:S1→S2\varphi: S_1 \to S_2 is an isometry between two regular C3C^3 surfaces (meaning φ\varphi is a diffeomorphism preserving the length of every curve), then for every p∈S1p \in S_1,

KS1(p)=KS2(φ(p)).K_{S_1}(p) = K_{S_2}(\varphi(p)).

This follows by choosing a local parametrization x\mathbf{x} of S1S_1 around pp and setting xˉ=φ∘x\bar{\mathbf{x}} = \varphi \circ \mathbf{x} as the induced parametrization of S2S_2 around φ(p)\varphi(p); the length-preservation of φ\varphi translates exactly into E=Eˉ,F=Fˉ,G=GˉE = \bar E, F = \bar F, G = \bar G, so the theorem applies directly.

Worked examples

Example 1: The cylinder, computed two ways. Parametrize the cylinder of radius aa by

x(u,v)=(acos⁡u, asin⁡u, v),(u,v)∈(0,2π)×R.\mathbf{x}(u,v) = (a\cos u,\ a\sin u,\ v), \qquad (u,v) \in (0, 2\pi) \times \mathbb{R}.

Then xu=(−asin⁡u,acos⁡u,0)\mathbf{x}_u = (-a\sin u, a\cos u, 0), xv=(0,0,1)\mathbf{x}_v = (0,0,1), giving

E=a2,F=0,G=1,E = a^2, \quad F = 0, \quad G = 1,

constants — independent of u,vu, v. All partial derivatives of E,F,GE, F, G vanish, so in the intrinsic formula of Step 3, every Christoffel symbol built from these derivatives is zero (each Γijk\Gamma^k_{ij} is a combination of first derivatives of E,F,GE,F,G divided by EG−F2EG-F^2), and hence K=0K = 0 by the intrinsic formula, with no reference to how the cylinder sits in space. Check directly via the extrinsic definition: xuu=(−acos⁡u,−asin⁡u,0)\mathbf{x}_{uu} = (-a\cos u, -a\sin u, 0), xuv=(0,0,0)\mathbf{x}_{uv} = (0,0,0), xvv=(0,0,0)\mathbf{x}_{vv} = (0,0,0), and N=(cos⁡u,sin⁡u,0)\mathbf{N} = (\cos u, \sin u, 0) (outward unit normal, since xu×xv=(acos⁡u,asin⁡u,0)\mathbf{x}_u \times \mathbf{x}_v = (a\cos u, a\sin u, 0) has length aa). Then

L=xuu⋅N=−acos⁡2u−asin⁡2u=−a,M=0,N=0.L = \mathbf{x}_{uu}\cdot \mathbf{N} = -a\cos^2 u - a \sin^2 u = -a, \qquad M = 0, \qquad N = 0.

So K=(LN−M2)/(EG−F2)=(0−0)/a2=0K = (LN - M^2)/(EG-F^2) = (0 - 0)/a^2 = 0. Both computations agree: K≡0K \equiv 0, matching the plane, exactly as the intuition of "rolling without stretching" predicts. Note L=−a≠0L=-a \neq 0: the cylinder genuinely bends in space (mean curvature H=EN−2FM+GL2(EG−F2)=−a2a2=−12a≠0H = \frac{EN - 2FM+GL}{2(EG-F^2)} = \frac{-a}{2a^2} = -\frac{1}{2a} \neq 0), but this extrinsic bending is invisible to KK.

Example 2: A non-constant-curvature surface — the surface of revolution of z=ln⁡(cos⁡u)z = \ln(\cos u), i.e. a "pseudo-catenoid-like" strip; concretely we use the graph z=f(u)=13u3z = f(u) = \tfrac13 u^3 over a line, rotated. To keep the arithmetic explicit and asymmetric, take the surface of revolution generated by rotating the profile curve r=eur = e^{u} (radius as a function of height uu) about the zz-axis, for u∈(−1,1)u \in (-1, 1):

x(u,v)=(eucos⁡v, eusin⁡v, u),v∈(0,2π).\mathbf{x}(u,v) = (e^u \cos v,\ e^u \sin v,\ u), \qquad v \in (0, 2\pi).

Compute:

xu=(eucos⁡v,eusin⁡v,1),xv=(−eusin⁡v,eucos⁡v,0).\mathbf{x}_u = (e^u\cos v, e^u \sin v, 1), \qquad \mathbf{x}_v = (-e^u \sin v, e^u \cos v, 0).

So

E=e2ucos⁡2v+e2usin⁡2v+1=e2u+1,F=−e2ucos⁡vsin⁡v+e2usin⁡vcos⁡v=0,E = e^{2u}\cos^2 v + e^{2u}\sin^2 v + 1 = e^{2u} + 1, \qquad F = -e^{2u}\cos v \sin v + e^{2u}\sin v \cos v = 0,
G=e2usin⁡2v+e2ucos⁡2v=e2u.G = e^{2u}\sin^2 v + e^{2u}\cos^2 v = e^{2u}.

Note EE depends only on uu and GG depends only on uu too; F≡0F \equiv 0 (this is the standard fact that meridians and parallels of a surface of revolution are orthogonal). For an orthogonal parametrization (F=0F=0), the intrinsic Gauss formula simplifies to:

K=−12EG[∂∂u(GuEG)+∂∂v(EvEG)].K = -\frac{1}{2\sqrt{EG}}\left[ \frac{\partial}{\partial u}\left(\frac{G_u}{\sqrt{EG}}\right) + \frac{\partial}{\partial v}\left(\frac{E_v}{\sqrt{EG}}\right) \right].

Since Ev=0E_v = 0 here, the second term vanishes, leaving

K=−12EG ∂∂u(GuEG).K = -\frac{1}{2\sqrt{EG}}\,\frac{\partial}{\partial u}\left(\frac{G_u}{\sqrt{EG}}\right).

We have G=e2uG = e^{2u} so Gu=2e2uG_u = 2e^{2u}, and EG=(e2u+1)e2uEG = (e^{2u}+1)e^{2u}, so EG=eue2u+1\sqrt{EG} = e^u\sqrt{e^{2u}+1}. Thus

GuEG=2e2ueue2u+1=2eue2u+1.\frac{G_u}{\sqrt{EG}} = \frac{2e^{2u}}{e^u\sqrt{e^{2u}+1}} = \frac{2e^{u}}{\sqrt{e^{2u}+1}}.

Differentiate with respect to uu using the quotient rule, with numerator derivative 2eu2e^u and denominator e2u+1\sqrt{e^{2u}+1}, denominator-derivative =e2ue2u+1= \dfrac{e^{2u}}{\sqrt{e^{2u}+1}}:

∂∂u(2eue2u+1)=2eue2u+1−2eu⋅e2ue2u+1e2u+1=2eu(e2u+1)−2e3u(e2u+1)3/2=2eu(e2u+1)3/2.\frac{\partial}{\partial u}\left(\frac{2e^u}{\sqrt{e^{2u}+1}}\right) = \frac{2e^u \sqrt{e^{2u}+1} - 2e^u \cdot \frac{e^{2u}}{\sqrt{e^{2u}+1}}}{e^{2u}+1} = \frac{2e^u(e^{2u}+1) - 2e^{3u}}{(e^{2u}+1)^{3/2}} = \frac{2e^u}{(e^{2u}+1)^{3/2}}.

Therefore

K=−12eue2u+1⋅2eu(e2u+1)3/2=−1(e2u+1)2.K = -\frac{1}{2 e^u \sqrt{e^{2u}+1}} \cdot \frac{2e^u}{(e^{2u}+1)^{3/2}} = -\frac{1}{(e^{2u}+1)^2}.

So KK is strictly negative and varies with uu — e.g. at u=0u=0: K(0)=−14K(0) = -\frac{1}{4}; at u=0.5u = 0.5: e1=2.71828…e^{1}=2.71828\ldots, (e2u+1)2=(2.71828+1)2=(3.71828)2=13.8256(e^{2u}+1)^2 = (2.71828+1)^2 = (3.71828)^2 = 13.8256, so K(0.5)≈−0.0723K(0.5) \approx -0.0723. This entire computation used only E,F,GE, F, G and their uu-derivatives — never the embedding data L,M,NL, M, N — exactly as the Theorema Egregium guarantees is possible.

Counterexample: why C3C^3 smoothness cannot be dropped

CounterexampleA merely C1C^1 "surface" for which mixed partials fail to commute, breaking the Gauss equation

The proof's Step 3 hinges entirely on the equality of mixed third-order partial derivatives xuuv=xuvu\mathbf{x}_{uuv} = \mathbf{x}_{uvu}, which is Clairaut/Schwarz's theorem, valid precisely because x\mathbf{x} is assumed C3C^3 (so that the second partials xuu,xuv,xvv\mathbf{x}_{uu}, \mathbf{x}_{uv}, \mathbf{x}_{vv} are C1C^1, and Schwarz's theorem on any C2C^2 function guarantees symmetric mixed second partials — here applied one derivative up, to xu\mathbf{x}_u itself being C2C^2). If instead we only assume x\mathbf{x} is C2C^2 but not C3C^3 — or artificially patch together a map whose second partial derivatives exist everywhere but are discontinuous — the equality of mixed third partials can fail, and the entire derivation of the Gauss equation collapses, since it is nothing but bookkeeping the identity xuuv=xuvu\mathbf{x}_{uuv} = \mathbf{x}_{uvu} in the frame {xu,xv,N}\{\mathbf{x}_u, \mathbf{x}_v, \mathbf{N}\}. To see this failure is not merely hypothetical, consider the classical example of a function whose mixed partials are unequal at a point, adapted to a graph. Define g:R2→Rg: \mathbb{R}^2 \to \mathbb{R} by

g(u,v)={uvu2−v2u2+v2(u,v)≠(0,0)0(u,v)=(0,0).g(u,v) = \begin{cases} uv\dfrac{u^2 - v^2}{u^2+v^2} & (u,v) \neq (0,0) \\ 0 & (u,v) = (0,0).\end{cases}

This function is C1C^1 everywhere and has second partial derivatives everywhere, but the classical computation shows

guv(0,0)=1whilegvu(0,0)=−1,g_{uv}(0,0) = 1 \qquad \text{while} \qquad g_{vu}(0,0) = -1,

so mixed partials disagree at the origin — precisely because gg fails to be C2C^2 there (its second partials are discontinuous at the origin, so Schwarz's theorem, which requires continuity of the mixed partial to guarantee equality, does not apply). Now consider the graph "surface" x(u,v)=(u,v,g(u,v))\mathbf{x}(u,v) = (u, v, g(u,v)). It is a perfectly good regular parametrized surface away from the origin, and it is even C1C^1 globally, but it fails to be C2C^2 (let alone C3C^3) at the origin. If one attempted to run the Step 3 argument through the origin, the derivation of the Gauss equation would require equating xuuv\mathbf{x}_{uuv} and xuvu\mathbf{x}_{uvu} there — but the very same failure of Schwarz's theorem that breaks guv(0,0)≠gvu(0,0)g_{uv}(0,0) \neq g_{vu}(0,0) propagates into the third-derivative identity underlying Gauss's argument, since the relevant second partials of x\mathbf{x} inherit gg's discontinuity. The compatibility identity that produces the intrinsic formula for KK is not available at such a point, and there is no guarantee that an extrinsically-computed KK (from L,M,NL, M, N, which still make sense away from any degeneracy) will match any purported "intrinsic formula," because that formula was never validly derived there. This shows precisely where smoothness is used: not in defining KK itself (which only needs C2C^2 to make sense as LN−M2EG−F2\frac{LN-M^2}{EG-F^2}), but in guaranteeing that KK can be re-expressed purely intrinsically — that step is where the extra derivative of regularity is essential, and dropping it can produce points where the intrinsic-formula derivation simply does not apply.

Scope, edge cases & common traps

SituationWhat holdsWhat can fail
x\mathbf{x} is C3C^3 (or C∞C^\infty, real-analytic)Theorema Egregium holds fully; KK intrinsic—
x\mathbf{x} is only C1C^1 or has a non-C2C^2 pointKK via L,M,NL,M,N may not even be defined, or the Gauss-equation derivation breaksIntrinsic formula for KK may not agree with extrinsic one
Two surfaces have same KK everywhere—They need not be locally isometric (converse of Theorema Egregium is false in general — e.g. a plane and a helicoid-esque surface can share K≡0K\equiv 0 pointwise without an isometry existing between arbitrary matching neighborhoods unless further conditions like matching principal curvature structure hold)
Global vs. localTheorem gives pointwise, local equality of KK under local isometryDoes not by itself give global topological consequences — that is the role of Gauss-Bonnet Theorem
Sign/orientation conventions for N\mathbf{N}KK is independent of the choice of unit normal orientation (both L,M,NL,M,N flip sign together, LN−M2LN-M^2 unchanged)Mean curvature HH does flip sign under N→−N\mathbf{N} \to -\mathbf{N} — do not confuse the two

The mathematician behind it

Origins · 1 minCarl Friedrich Gauss and how this was discovered
CF
Carl Friedrich Gauss
1777–1855 · Göttingen, Kingdom of Hanover

Gauss was one of history's most versatile mathematicians, contributing foundational results in number theory, algebra, statistics, astronomy, and geodesy. His work on differential geometry grew directly out of his practical work surveying the Kingdom of Hanover in the 1820s, where he needed to understand how measurements taken on the curved surface of the Earth related to the underlying geometry — precisely the question of what can be known "intrinsically." His 1827 treatise Disquisitiones Generales circa Superficies Curvas ("General Investigations of Curved Surfaces") introduced the curvature that now bears his name and proved the Theorema Egregium, a result he considered among his most beautiful discoveries — hence his own choice of the adjective "remarkable" (egregium) in naming it.

History
- 1777: Gauss born in Braunschweig.
- 1818–1825: Gauss directs the geodetic survey of the Kingdom of Hanover, motivating deep questions about curved-surface measurement.
- 1827: Disquisitiones Generales circa Superficies Curvas published, containing the Theorema Egregium.
- 1854: Bernhard Riemann, Gauss's student, delivers his habilitation lecture generalizing curvature to arbitrary dimensions, founding Riemannian geometry.
- 1855: Gauss dies in Göttingen.
Does the Theorema Egregium mean the second fundamental form is useless?

No. The second fundamental form still encodes essential extrinsic information — how the surface actually sits in R3\mathbb{R}^3, including mean curvature HH and the individual principal curvatures κ1,κ2\kappa_1, \kappa_2 (with K=κ1κ2K = \kappa_1\kappa_2, H=(κ1+κ2)/2H = (\kappa_1+\kappa_2)/2). The theorem shows only that the specific product κ1κ2\kappa_1\kappa_2 is secretly intrinsic; HH is not intrinsic (a cylinder and a plane share K=0K=0 but have different HH: −1/(2a)-1/(2a) versus 00). So the second fundamental form remains necessary for questions about how a surface bends in space, minimal surface theory, and rigidity — the theorem isolates exactly one combination of curvatures as intrinsic, not the whole extrinsic package.

Is the converse true: if two surfaces have the same Gaussian curvature function, must they be locally isometric?

Not in general. Matching KK pointwise is necessary but not automatically sufficient for local isometry; genuine converse-type theorems (e.g. surfaces of constant curvature KK are locally isometric to the sphere, plane, or hyperbolic plane depending on the sign of KK, by Minding's theorem) require the curvature to be constant, plus care about orientation and completeness. For non-constant curvature functions, having matching K(u,v)K(u,v) at corresponding points of some parametrization does not by itself construct an isometry; the full converse direction is a separate, harder theorem with its own hypotheses, and should not be assumed from the Theorema Egregium alone.

Why is it called "remarkable" — what exactly surprised Gauss?

The surprise is structural: the definition of K=(LN−M2)/(EG−F2)K = (LN-M^2)/(EG-F^2) manifestly uses L,M,NL, M, N, which are built from the second derivatives of the embedding x\mathbf{x} and the normal vector N\mathbf{N} — quantities that seem to fundamentally require knowing the surface's position in R3\mathbb{R}^3. Gauss discovered that when you actually carry out the algebra of comparing mixed partial derivatives, all dependence on L,M,NL, M, N individually cancels out, leaving a formula in E,F,GE, F, G alone. There was no a priori reason to expect this; it emerges only from the specific compatibility identity forced by equality of mixed partials.

How does this relate to the Gauss-Bonnet theorem?

The Theorema Egregium is local and purely metric: it shows KK is computable from the first fundamental form at each point. The Gauss-Bonnet Theorem takes this pointwise-defined intrinsic quantity and integrates it over an entire closed surface, proving ∬SK dA=2πχ(S)\iint_S K \, dA = 2\pi \chi(S), where χ(S)\chi(S) is the Euler characteristic — a purely topological invariant. So the Theorema Egregium supplies the object (KK, known to be intrinsic) that Gauss-Bonnet then reveals has an unexpectedly rigid global, topological meaning, bridging differential geometry and topology.

Can Gaussian curvature be negative, and what does that mean intrinsically?

Yes — surfaces like the hyperbolic paraboloid (saddle) or the pseudosphere have K<0K < 0 at points where the surface curves away from its tangent plane in two opposite senses (one principal curvature positive, one negative). Intrinsically, negative curvature manifests as triangles whose angle sums are less than π\pi, and as exponential divergence of nearby geodesics — features an ant confined to the surface could detect using a protractor and a notion of "straightest paths," with no need to see the embedding, exactly as the Theorema Egregium guarantees.

What role does the requirement x\mathbf{x} be C3C^3 actually play, precisely?

It guarantees that the second partial derivatives of x\mathbf{x} are themselves C1C^1, so that Schwarz's theorem on equality of mixed partials applies to them, giving xuuv=xuvu\mathbf{x}_{uuv} = \mathbf{x}_{uvu}. This single identity is the entire engine of the proof: expanding both sides in the frame {xu,xv,N}\{\mathbf{x}_u,\mathbf{x}_v,\mathbf{N}\} and matching coefficients produces the Gauss equation, from which L,M,NL,M,N-dependence cancels. Without this smoothness, that cancellation is not guaranteed to occur, and the derivation of an intrinsic formula for KK has no logical foundation at points of insufficient regularity.

Does the theorem apply to abstract Riemannian manifolds, not just surfaces embedded in R3\mathbb{R}^3?

Yes, and this generalization is exactly Riemann's contribution. Given any smooth manifold with a Riemannian metric (not assumed to arise from any embedding), one can still define an intrinsic curvature (the sectional curvature in dimension 2 reduces to Gauss's KK) using only the metric's coefficients and their derivatives, via the identical Christoffel symbol machinery from this proof. The Theorema Egregium is thus not merely a fact about surfaces sitting in R3\mathbb{R}^3; it is the historical origin of defining curvature intrinsically for any metric space of this type, embedded or not.

Before this

Where this leads